WO2019001359A1 - 数据处理方法和数据处理装置 - Google Patents

数据处理方法和数据处理装置 Download PDF

Info

Publication number
WO2019001359A1
WO2019001359A1 PCT/CN2018/092390 CN2018092390W WO2019001359A1 WO 2019001359 A1 WO2019001359 A1 WO 2019001359A1 CN 2018092390 W CN2018092390 W CN 2018092390W WO 2019001359 A1 WO2019001359 A1 WO 2019001359A1
Authority
WO
WIPO (PCT)
Prior art keywords
incremental
decision tree
model
data
data processing
Prior art date
Application number
PCT/CN2018/092390
Other languages
English (en)
French (fr)
Inventor
宋文鹏
沈雄
Original Assignee
众安信息技术服务有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 众安信息技术服务有限公司 filed Critical 众安信息技术服务有限公司
Priority to KR1020197013526A priority Critical patent/KR20190075962A/ko
Publication of WO2019001359A1 publication Critical patent/WO2019001359A1/zh
Priority to US16/362,186 priority patent/US20190220710A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G06F18/2193—Validation; Performance evaluation; Active pattern learning techniques based on specific statistical tests
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • the invention belongs to the field of data processing, and in particular relates to a data processing method and a data processing device.
  • the random forest classification model is one of the more commonly used classification models.
  • the classification model consists of multiple decision trees. When the samples to be classified enter the random forest, the multiple decision trees are classified, and finally the selection times of all decision trees are selected. The most categories are the final classification results.
  • the offline machine learning process is usually used to construct the classification model. Through the learning, analysis and training of the full amount of user behavior data, the knowledge about the classification is obtained, thereby completing the construction of the classification model and deploying the online. Over time, the classification models deployed on-line will generally degrade and the accuracy of their classification may not be sufficient.
  • the embodiment of the present invention provides a data processing method and a data processing apparatus to solve the problem that the existing prediction models are all offline prediction modes, and adaptive update cannot be implemented.
  • an embodiment of the present invention provides a data processing method, where the data processing method includes generating at least one incremental decision tree according to incremental data; and based on a plurality of model decision trees and at least one incremental decision tree pair in the classification model The incremental data is predicted and the predicted result is obtained; the classification model is updated according to the predicted result.
  • generating at least one incremental decision tree according to the incremental data comprising: extracting a plurality of sample sets based on the incremental data; generating at least one incremental decision tree based on the plurality of sample sets, wherein The number of decision trees is determined based on the number of model decision trees.
  • the classification model is updated according to the prediction result, including obtaining comprehensive performance of at least one incremental decision tree and multiple model decision trees according to the prediction result; and integrating based on the incremental decision tree and the model decision tree Performance, selecting a predetermined number of decision trees from a plurality of model decision trees and at least one incremental decision tree as a model decision tree in the updated classification model.
  • the predetermined number is equal to the number of plurality of model decision trees.
  • the comprehensive performance of the at least one incremental decision tree and the plurality of model decision trees is obtained according to the prediction result, including establishing time and incrementing based on the at least one incremental decision tree and the plurality of model decision trees.
  • the prediction accuracy of the data determines the overall performance.
  • the incremental data is predicted based on the plurality of model decision trees and the at least one incremental decision tree in the classification model, including multiple model decision trees and at least one incremental decision tree based on the classification model Label prediction for incremental data.
  • the method further includes determining, according to a result of the label prediction, a prediction accuracy of the plurality of model decision trees and the at least one incremental decision tree on the incremental data; and the plurality of model decision trees and the at least one incremental decision
  • the establishment time of the tree is used as the weight for determining the comprehensive performance, and the prediction accuracy of the incremental data is sorted.
  • the weight of the decision tree with long establishment time is smaller than the weight of the decision tree with short establishment time.
  • the number of incremental decision trees is determined based on the number of model decision trees.
  • the number of incremental decision trees is equal to 10% to 30% of the number of model decision trees.
  • the method further includes: acquiring incremental data in a predetermined time period, and determining, according to whether a classification model exists, determining a quantity of generating at least one incremental decision tree; wherein, if there is a classification model, generating according to the incremental data At least one incremental decision tree.
  • the method further includes: if there is no classification model, creating a classification model including a plurality of model decision trees according to the historical data, wherein the historical data is classified data.
  • the data processing method includes acquiring incremental data within a predetermined time period, and determining a number of generated decision trees based on whether a classification model exists; and if there is a classification model, generating an increment according to the incremental data Decision tree, and based on the incremental decision tree and the model decision tree and the incremental decision tree in the classification model, the incremental data is tagged, wherein the number of incremental decision trees is determined based on the number of model decision trees before the update.
  • an embodiment of the present invention further provides a data processing apparatus, where the data processing apparatus includes an incremental decision tree generating module, configured to generate at least one incremental decision tree according to the incremental data, and a prediction module, configured to be based on the classification model The plurality of model decision trees and the at least one incremental decision tree predict the incremental data and obtain the predicted result; and the update module is configured to update the classification model according to the predicted result.
  • the data processing apparatus includes an incremental decision tree generating module, configured to generate at least one incremental decision tree according to the incremental data, and a prediction module, configured to be based on the classification model The plurality of model decision trees and the at least one incremental decision tree predict the incremental data and obtain the predicted result; and the update module is configured to update the classification model according to the predicted result.
  • the incremental decision tree generating module includes a sampling unit configured to extract a plurality of sample sets based on the incremental data, and a generating unit configured to generate at least one incremental decision based on the plurality of sample sets A tree, wherein the number of at least one incremental decision tree is determined based on the number of multiple model decision trees.
  • the update module includes an integrated performance determining unit, configured to obtain a comprehensive performance of the at least one incremental decision tree and the plurality of model decision trees according to the prediction result; and the updating unit is configured to perform the at least one incremental decision
  • the comprehensive performance of the tree and the plurality of model decision trees selects a predetermined number of decision trees from the plurality of model decision trees and the at least one incremental decision tree as the model decision tree in the updated classification model.
  • the data processing apparatus includes: an incremental data input unit configured to acquire incremental data within a predetermined time period; and a determining unit configured to generate a representation of the existing classification model according to whether a classification model exists a first signal and a second signal characterizing the absence of the classification model; the decision tree generation unit configured to generate an incremental decision tree based on the incremental data based on the first signal of the response; the label prediction unit configured to be in accordance with the classification model a model decision tree and an incremental decision tree for tag prediction of incremental data; a decision tree selection unit configured to select a predetermined performance based on a comprehensive performance of each of the decision trees in the model decision tree and the incremental decision tree A number of decision trees; a model update unit configured to use the selected predetermined number of decision trees as model decision trees in the updated classification model.
  • the predetermined number in the update unit is equal to the number of the plurality of model decision trees.
  • the comprehensive performance determining unit is further configured to determine the comprehensive performance based on the setup time of the at least one incremental decision tree and the plurality of model decision trees and the prediction accuracy rate for the incremental data.
  • the prediction module is configured to perform tag prediction on the incremental data based on the plurality of model decision trees and the at least one incremental decision tree in the classification model.
  • the prediction module is further configured to determine, according to a result of the label prediction, a prediction accuracy of the plurality of model decision trees and the at least one incremental decision tree on the incremental data; and the plurality of model decision trees and at least one The establishment time of the incremental decision tree is used as the weight for determining the comprehensive performance, and the prediction accuracy of the incremental data is sorted.
  • the weight of the decision tree with long establishment time is smaller than the weight of the decision tree with short establishment time.
  • the number of at least one incremental decision tree in the incremental decision tree generation module is determined according to the number of the plurality of model decision trees.
  • the number of at least one incremental decision tree in the incremental decision tree generation module is equal to 10% to 30% of the number of the plurality of model decision trees.
  • the incremental decision tree generating module is further configured to acquire incremental data within a predetermined time period, and determine, according to whether a classification model exists, determine the number of generated at least one incremental decision tree; wherein, if there is a classification The model generates at least one incremental decision tree based on the incremental data.
  • the incremental decision tree generating module is further configured to: if there is no classification model, create a classification model including a plurality of model decision trees according to the historical data, wherein the historical data is classified data.
  • the embodiment of the present invention further provides a computer storage medium, where the data processing program is stored, and the data processing program is executed by the processor to implement data processing mentioned in any of the above embodiments. The operation of the method.
  • the data processing method provided by the embodiment of the invention updates the classification model by using the incremental data, so that the classification model can make corresponding adjustments according to the changes of the sample data in a timely or near real-time manner, and realizes the classification model and the latest sample data. Synchronize. That is to say, the data processing method provided by the embodiment of the present invention can perform adaptive update based on the currently newly obtained data, thereby adapting to the new trend change of the data, thereby ensuring the accuracy of the prediction.
  • the embodiment of the present invention achieves the purpose of eliminating the need for manual intervention in the service cycle of the model, and greatly saves the cost, so that the data processing method provided by the embodiment of the present invention is intelligent and efficient. specialty.
  • FIG. 1 is a schematic flowchart diagram of a data processing method according to an embodiment of the present invention.
  • FIG. 2 is a schematic flowchart of generating at least one incremental decision tree operation according to incremental data according to a data processing method according to an embodiment of the present invention.
  • FIG. 3 is a schematic flowchart diagram of an update operation of a classification model according to a prediction result according to a data processing method according to an embodiment of the present invention.
  • FIG. 4 is a schematic flowchart diagram of a data processing method according to another embodiment of the present invention.
  • FIG. 5 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present invention.
  • FIG. 6 is a schematic structural diagram of an incremental decision tree generating module of a data processing apparatus according to an embodiment of the present invention.
  • FIG. 7 is a schematic structural diagram of an update module of a data processing apparatus according to an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of a data processing apparatus according to another embodiment of the present invention.
  • FIG. 9 is a schematic structural diagram of a decision tree selection unit of a data processing apparatus according to an embodiment of the present invention.
  • FIG. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
  • the embodiment of the present invention proposes a technical solution for generating an incremental decision tree based on incremental data and then updating the classification model.
  • the incremental data can be from financial product information transmitted via a network, such as price, transaction amount, transaction volume, and the like.
  • the random forest classification model is a classifier containing multiple decision trees, and the output classification result is determined by the total number of classification results output by a single decision tree.
  • the basic idea of random forest classification is to randomly extract N sample sets from the original sample set, and the sample size of each sample set is the same as the original sample set; N is established for each N sample set.
  • Decision trees each decision tree has a vote option to select the classification results, and obtain N classification results; each sample is voted according to the N classification results to determine its final classification.
  • the process of random forest generation is the process of training each decision tree.
  • the process of training each decision tree includes the following: (1) randomly selecting M samples with a return, and training a decision tree with the M samples; (2) each sample has multiple attributes in the decision tree When a split node is needed, m attributes are randomly selected from the plurality of attributes, and then a specific attribute is selected from the m attributes to select the best attribute as the split attribute of the current node; (3) each node of the decision tree The splits are carried out according to (2) until they cannot be split.
  • the classification model deployed on the line that is, the classification model consisting of a predetermined number of model decision trees
  • the classification model deployed on the line can be used to perform category prediction by scoring, and the category with the highest score will be used. (Select the largest number of decision trees in this category) as the forecast category, and based on the forecast category, carry out pre-set business applications, such as determining the price rise and fall by category.
  • FIG. 1 is a schematic flowchart diagram of a data processing method according to an embodiment of the present invention.
  • the method of Figure 1 is performed by a server or computing device.
  • the data processing method provided by the embodiment of the present invention includes the following content.
  • incremental data refers to new data acquired over a certain period of time (eg, 10 minutes, 1 hour, or 1 day) from a data storage device or server.
  • An incremental decision tree is a tree structure in which each internal node of an incremental decision tree represents an attribute test, each branch represents a test output, and each leaf node represents a category.
  • model decision tree is also a tree structure in which each internal node of the model decision tree represents an attribute test, each branch represents a test output, and each leaf node represents a category.
  • the prediction operation of the incremental data is performed by means of a label prediction method.
  • the incremental data is subjected to back-sampling, a certain number of sample sets are extracted, and then a corresponding number of incremental decision trees are generated based on the extracted sample sets, and finally, the incremental data is subjected to label prediction operations based on the generated incremental decision trees.
  • the prediction results should be able to reflect the overall performance of the incremental decision tree, especially for the prediction accuracy of incremental data.
  • the incremental decision tree is first generated based on the incremental data, and then the incremental data is predicted based on the model decision tree and the incremental decision tree to obtain the prediction result. Finally, the model decision tree in the classification model is based on the prediction result. Perform an update operation.
  • the update operation refers to selecting an incremental decision tree with better comprehensive performance to replace the model decision tree with poor comprehensive performance in the pre-update classification model.
  • the data processing method provided by the embodiment of the present invention generates an incremental decision tree by using incremental data, and then predicts the incremental data based on the model decision tree and the incremental decision tree in the classification model, and updates the classification model according to the prediction result. In this way, the adaptive update of the classification model is realized, and the purpose of no manual intervention is needed in the business cycle of the model, which greatly saves the cost.
  • FIG. 2 is a schematic flowchart of generating at least one incremental decision tree operation according to incremental data according to a data processing method according to an embodiment of the present invention.
  • generating at least one incremental decision tree (11) according to the incremental data includes:
  • the data processing method provided by the embodiment of the present invention generates an incremental decision tree by adopting a method of extracting multiple sample sets in a reversible manner, so that each node of the incremental decision tree is selected from the characteristics of the sample set.
  • the purpose is to provide a prerequisite for the final improvement of the prediction accuracy of the classification model.
  • FIG. 3 is a schematic flowchart diagram of an update operation of a classification model according to a prediction result according to a data processing method according to an embodiment of the present invention.
  • the classification model is updated according to the prediction result (13), including:
  • evaluation parameters in the comprehensive performance may be set according to actual conditions, including but not limited to evaluation parameters such as establishment time, prediction accuracy, and the like.
  • the data processing method provided by the embodiment of the present invention selectively updates the model decision tree in the pre-update classification model according to the comprehensive performance of each decision tree, and replaces the update with a comprehensive decision tree with better comprehensive performance.
  • the model decision tree in the former classification model which ultimately achieves accurate prediction of the updated classification model.
  • FIG. 4 is a schematic flowchart diagram of a data processing method according to another embodiment of the present invention. As shown in FIG. 4, the data processing method provided by the embodiment of the present invention includes the following content.
  • acquiring incremental data refers to acquiring incremental data for a predetermined period of time from a financial transaction server or a particular storage device.
  • the predetermined time period refers to a time period before the current time, and the length of the predetermined time period may be set according to specific requirements, as long as the user behavior data in the predetermined time period is already in an available state and already contains the actual
  • the category label information can be.
  • the length of the predetermined time period may be in days, in hours, or in minutes.
  • a financial product (such as stock) transaction is taken as an example for description.
  • the label of the data may be rising, falling, and flat, that is, the predetermined time period is a time period within 5 minutes before the current time.
  • the tags of the data may have many other forms.
  • Scenario 1 There is a classification model.
  • c represents the actual category of the sample.
  • a classification model is used to classify and predict the trend of stock prices, and the attributes of each sample may selectively include specific attributes such as stock name, price, and transaction volume.
  • K may be set according to the actual situation, so as to fully improve the adaptability and application of the data processing method provided by the embodiment of the present invention, which is not limited by the embodiment of the present invention.
  • each sample set grows into a corresponding delta decision tree, ie each node of the tree is a feature selected from the sample set.
  • the model decision tree (assumed T) and the K incremental decision trees in the classification model are used for label prediction (ie, classification prediction), and the unclassified incremental data is classified.
  • label prediction ie, classification prediction
  • the unclassified incremental data is classified.
  • T+K decision trees for tag prediction of incremental data. Because the total number of decision trees participating in the prediction increases and the K incremental decision trees often represent new trend changes, the use of T+K decision trees is beneficial to improve the accuracy of classification model prediction.
  • the value range of the setting K is 0.1T to 0.3T.
  • T, K are only used to characterize the model decision tree in the classification model and the number of incremental decision trees generated from the incremental data, and are not intended to limit T, K to a specific value, such as greater than or An integer equal to 1.
  • the prediction result is first obtained based on the label prediction operation performed in 45, and then the prediction result is compared with the real result, thereby obtaining the current accuracy of each decision tree, that is, the prediction accuracy for the incremental data. rate. Accordingly, the settling time of each decision tree can also be obtained, that is, the time that each decision tree already exists.
  • the comprehensive performance a * setup time + b * prediction accuracy, wherein a, b are the weights of the setup time and the prediction accuracy, respectively, and the values of a and b can be adjusted according to actual conditions.
  • the generation time of the decision tree also affects the comprehensive performance, that is, the weight of the decision tree closest to the current time is greater than the weight of the decision tree that is longer than the current time.
  • the values of a and b it is possible to make the decision tree with shorter settling time better than the decision with longer settling time when the prediction accuracy of the two decision trees is the same.
  • the overall performance of the tree is configured.
  • the setup time is introduced as a weight that affects the overall performance of the decision tree.
  • the two decision trees are further determined according to the establishment time of the two decision trees.
  • the comprehensive performance that is, due to the short setup time of decision tree 4, results in the conclusion that the overall performance of decision tree 4 is better than the comprehensive performance of decision tree 5.
  • a predetermined number of decision trees are selected as the model decision tree of the updated classification model, wherein the comprehensive performance ranking of each decision tree is based on the label prediction of the incremental data by each decision tree.
  • the result is. Specifically, the decision tree is sorted based on the comprehensive performance of the decision tree to obtain a decision tree sequence sorted according to the comprehensive performance shown in Table 1, and a decision tree with excellent comprehensive performance is selected according to the sort result. It can be seen from the foregoing that when considering the weight of the setup time, the overall performance of the decision tree 4 will be better than the comprehensive performance of the decision tree 5, so if four decision trees are required to discard one decision tree, the decision tree 5 will be discarded.
  • the decision trees 1 to 4 will be selected as the model decision tree of the classification model, and the updated classification model will be used to predict the subsequent incremental data.
  • the data processing method provided by the embodiment of the present invention can implement the update operation of the classification model under the premise of ensuring the prediction accuracy of the classification model.
  • the number K of incremental decision trees is determined based on the number T of model decision trees in the classification model.
  • the number K of incremental decision trees ranges from 10% to 30% of the number T of model decision trees in the classification model. Further, the specific value of K can be randomly determined between 10% and 30% of T according to the user's instruction or application scenario, so that the number T of model decision trees in the classification model can also produce corresponding changes. It should be understood that the limitation of the number of incremental decision trees in the embodiment of the present invention achieves the purpose of not affecting the stability of the classification model in the case of updating the classification model.
  • the number of selected predetermined number of decision trees is equal to the number of original model decision trees in the classification model, that is, the number of model decision trees in the classification model is always kept as T, and discarded.
  • the number of decision trees is equal to the number of incremental decision trees.
  • the incremental data is subjected to label prediction using T+K (ie, 240) decision trees, and then the comprehensive performance of the decision tree is sorted based on the prediction result.
  • T+K ie, 240
  • 190, 200 or 210 decision trees can be selected from the 240 decision trees as the model decision tree of the classification model, thereby completing the updating of the classification model.
  • K may be any one of 0.1T to 0.3T or a user-specified number when updating with the classification model next time.
  • Scenario 2 There is no classification model.
  • a model decision tree is generated based on the historical data, for example, the historical data is sampled to form T sample sets, and then based on the T The sample set generates T model decision trees. It can be understood that historical data is classified data.
  • the embodiment of the present invention does not adopt a traditional offline calculation method for reconstructing a classification model based on full-quantity data, but uses incremental data to update the classification model, so that the classification model can be made according to the change of the sample data in a timely or near real-time manner. Corresponding adjustments enable synchronization of the classification model with the latest sample data.
  • the embodiment of the present invention achieves the purpose of eliminating the need for manual intervention in the service cycle of the model, and greatly saves the cost, so that the data processing method provided by the embodiment of the present invention is intelligent and efficient. specialty.
  • FIG. 5 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present invention. As shown in FIG. 5, the data processing apparatus provided by the embodiment of the present invention includes:
  • the incremental decision tree generating module 51 is configured to generate at least one incremental decision tree according to the incremental data.
  • the prediction module 52 is configured to predict the incremental data based on the plurality of model decision trees and the at least one incremental decision tree in the classification model to obtain the prediction result.
  • the updating module 53 is configured to update the classification model according to the prediction result.
  • the prediction module 52 is configured to perform label prediction on the incremental data based on the plurality of model decision trees and the at least one incremental decision tree in the classification model.
  • the prediction module 52 is further configured to determine, according to a result of the label prediction, a prediction accuracy of the plurality of model decision trees and the at least one incremental decision tree on the incremental data;
  • the establishment time of at least one incremental decision tree is used as a weight for determining the comprehensive performance, and the prediction accuracy of the incremental data is ranked, wherein the weight of the decision tree with a long establishment time is smaller than the weight of the decision tree with a short establishment time.
  • the number of at least one incremental decision tree in the incremental decision tree generation module 51 is determined according to the number of multiple model decision trees.
  • the number of at least one incremental decision tree in the incremental decision tree generation module 51 is equal to 10% to 30% of the number of the plurality of model decision trees.
  • the incremental decision tree generating module 51 is further configured to acquire incremental data within a predetermined time period, and determine, according to whether a classification model exists, the number of generated at least one incremental decision tree; There is a classification model that generates at least one incremental decision tree based on the incremental data.
  • the incremental decision tree generating module 51 is further configured to create a classification model including a plurality of model decision trees according to historical data if there is no classification model, wherein the historical data is classified data.
  • FIG. 6 is a schematic structural diagram of an incremental decision tree generating module of a data processing apparatus according to an embodiment of the present invention.
  • the incremental decision tree generating module 51 of the data processing apparatus provided by the embodiment of the present invention includes:
  • the sampling unit 61 is configured to extract a plurality of sample sets based on the incremental data.
  • the generating unit 62 is configured to generate at least one incremental decision tree based on the plurality of sample sets, wherein the number of the at least one incremental decision tree is determined based on the number of the plurality of model decision trees.
  • FIG. 7 is a schematic structural diagram of an update module of a data processing apparatus according to an embodiment of the present invention.
  • the update module 53 of the data processing apparatus provided by the embodiment of the present invention includes:
  • the comprehensive performance determining unit 71 is configured to obtain comprehensive performance of at least one incremental decision tree and multiple model decision trees according to the prediction result.
  • the updating unit 72 is configured to select, according to the comprehensive performance of the at least one incremental decision tree and the plurality of model decision trees, a predetermined number of decision trees from the plurality of model decision trees and the at least one incremental decision tree as the updated classification model Model decision tree.
  • the predetermined number in the update unit 72 is equal to the number of multiple model decision trees.
  • the comprehensive performance determining unit 71 is further configured to determine the comprehensive performance based on the setup time of the at least one incremental decision tree and the plurality of model decision trees and the prediction accuracy rate for the incremental data.
  • FIG. 8 is a schematic structural diagram of a data processing apparatus according to another embodiment of the present invention. As shown in FIG. 8, the data processing apparatus provided by the embodiment of the present invention includes:
  • the incremental data input unit 81 is configured to acquire incremental data within a predetermined period of time.
  • the determining unit 82 is configured to generate a first signal representing the presence of the classification model and a second signal characterizing the absence of the classification model according to whether there is a classification model.
  • the decision tree generation unit 83 is configured to generate an incremental decision tree based on the incremental data based on the first signal.
  • the tag prediction unit 84 is configured to perform tag prediction on the delta data according to the model decision tree and the delta decision tree in the classification model.
  • the decision tree selection unit 85 is configured to select a predetermined number of decision trees based on the overall performance of the model decision trees in the classification model and the individual decision trees in the incremental decision tree.
  • the model update unit 86 is configured to use the selected predetermined number of decision trees as a model decision tree in the updated classification model.
  • the data processing apparatus provided by the embodiment of the present invention can predict the incremental data by using a classification model after acquiring the incremental data, and can also update the classification model based on the incremental data. That is to say, the data processing apparatus provided by the embodiment of the present invention implements adaptive updating of the classification model.
  • the number of predetermined number of decision trees selected by decision tree selection unit 85 is equal to the number of original model decision trees in the classification model.
  • the data processing apparatus further includes a historical data input unit 87 configured to acquire the classified historical data. Specifically, when the determination unit 82 does not find a classification model that can be used, the decision tree generation unit 83 generates a model decision tree based on the history data based on the second signal generated by the determination unit 82, thereby generating a classification model that can be used.
  • FIG. 9 is a schematic structural diagram of a decision tree selection unit of a data processing apparatus according to an embodiment of the present invention.
  • the decision tree selecting unit 85 includes an accuracy determining unit 91 and a decision tree comprehensive performance sorting unit 92, wherein the accuracy determining unit 91 is configured to be based on the label.
  • the result of the prediction determines the prediction accuracy of each decision tree for incremental data
  • the decision tree comprehensive performance ranking unit 92 is configured to sort based on the setup time of each decision tree and the prediction accuracy of the incremental data; wherein, the setup time
  • the weight of a long decision tree is less than the weight of a decision tree with a short build time.
  • the operations and functions of the integrated performance determining unit 71 and the updating unit 72 included in the update module 53 may refer to the data processing methods provided in the above-mentioned FIG. 1 to FIG. 4, and are not described herein again in order to avoid redundancy.
  • FIG. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
  • the electronic device provided in FIG. 10 is for performing the data processing method described in the embodiments of FIGS. 1 through 4.
  • the electronic device includes a processor 101, a memory 102, and a bus 103.
  • the processor 101 is configured to invoke, by using the bus 103, the code stored in the memory 102 to generate at least one incremental decision tree according to the incremental data; based on the plurality of model decision trees in the classification model and the at least one incremental decision tree pair increment
  • the data is predicted and the predicted results are obtained; the classification model is updated based on the predicted results.
  • the electronic device includes, but is not limited to, an electronic device such as a mobile phone or a tablet computer.
  • a computer storage medium is further provided, where the data processing program is stored, and the data processing program is executed by the processor to implement the data processing mentioned in any of the above embodiments. The operation of the method.
  • the computer readable medium is a CD-ROM, a floppy disk, a hard disk, a digital versatile disk (DVD), a Blu-ray disk or other form of memory.
  • some or all of the example methods of FIGS. 1-4 may utilize an application specific integrated circuit (ASIC), a programmable logic device (PLD), an on-site programmable logic device (EPLD), discrete logic, hardware, Any combination of firmware and the like is implemented.
  • ASIC application specific integrated circuit
  • PLD programmable logic device
  • EPLD on-site programmable logic device
  • FIGS. 1 to 4 describes the data processing method, the operations in the processing method may be modified, deleted, or merged.
  • any of Figures 1 through 4 can be implemented with encoded instructions (such as computer readable instructions) stored on a tangible computer readable medium, such as a hard disk, flash memory, read only memory (ROM) ), a compact disc (CD), a digital versatile disc (DVD), a cache, a random access memory (RAM), and/or any other storage medium on which information can be stored for any time (eg, for a long time, permanently , short-lived situations, temporary buffering, and/or caching of information).
  • a tangible computer readable medium is expressly defined to include any type of computer readable stored signal. Additionally or alternatively, the example process of FIG.
  • 1 may be implemented with encoded instructions (such as computer readable instructions) stored on a non-transitory computer readable medium such as a hard disk, flash memory, read only memory, optical disk, digital general purpose An optical disc, a cache, a random access memory, and/or any other storage medium in which information can be stored at any time (eg, for a long time, permanently, transiently, temporarily buffered, and/or cached of information).
  • a non-transitory computer readable medium such as a hard disk, flash memory, read only memory, optical disk, digital general purpose An optical disc, a cache, a random access memory, and/or any other storage medium in which information can be stored at any time (eg, for a long time, permanently, transiently, temporarily buffered, and/or cached of information).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Software Systems (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种数据处理方法和数据处理装置，该数据处理方法包括根据增量数据生成至少一个增量决策树；基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测并获得预测结果；根据预测结果对分类模型进行更新。本发明实施例提供的数据处理方法通过利用增量数据生成增量决策树，然后基于分类模型中的模型决策树和增量决策树对增量数据进行预测，根据预测结果对分类模型进行更新的方式，实现了分类模型的自适应更新，并且实现了在模型的业务周期内不再需要人工干预的目的，大大节约了成本。

Description

数据处理方法和数据处理装置
本申请要求2017年06月30日提交的申请号为No.201710523102.5的中国申请的优先权，通过引用将其全部内容并入本文。
技术领域
本发明属于数据处理领域，尤其涉及一种数据处理方法和数据处理装置。
发明背景
随着互联网技术的发展，出现了大量的网络应用，例如：网络社交、网络阅读、股票基金交易等等。网络应用提供方为了向用户推荐具有针对性的信息，通常会周期性地对当前的数据进行处理，然后，向用户推送预测性的信息。为了提高预测效率和准确性，在大多数网络应用中通常采用分类模型进行分类预测。
随机森林分类模型是应用比较普遍的分类模型之一，该分类模型由多棵决策树组成，当待分类样本进入随机森林时，由该多棵决策树进行分类，最后选取被所有决策树选择次数最多的类别作为最终的分类结果。在传统应用中，通常采用离线的机器学习过程构造该分类模型，通过对全量的用户行为数据的学习、分析与训练，得出关于分类的知识，从而完成对分类模型的构建并部署上线。随着时间的推移，在线上部署的分类模型通常会逐渐退化，其分类的准确率可能无法满足要求。
在传统的机器学习领域，一般都是基于离线学习的机器学习模型。然而，随着数据量的增大，机器学习模型的处理能力越来越下降，特别在金融交易领域，信息瞬息万变，离线的机器学习模型会导致交易系统出现一定程度的滞后性。
因此，亟需一种能够自动进行更新的预测模型对数据进行处理。
发明内容
有鉴于此，本发明实施例提出一种数据处理方法和数据处理装置，以解决现有预测模型均为离线预测模式，无法实现自适应更新的问题。
第一方面，本发明实施例提供一种数据处理方法，该数据处理方法包括根据增量数据生成至少一个增量决策树；基于分类模型中的多个模型决 策树和至少一个增量决策树对增量数据进行预测并获得预测结果；根据预测结果对分类模型进行更新。
在本发明一实施例中，根据增量数据生成至少一个增量决策树，包括基于增量数据有放回地抽取多个样本集；基于多个样本集生成至少一个增量决策树，其中增量决策树的数量基于模型决策树的数量来确定。
在本发明一实施例中，根据预测结果对分类模型进行更新，包括根据预测结果得出至少一个增量决策树和多个模型决策树的综合性能；基于增量决策树和模型决策树的综合性能，从多个模型决策树和至少一个增量决策树中选取预定数量的决策树来作为更新后分类模型中的模型决策树。
在本发明一实施例中，预定数量等于多个模型决策树的数量。
在本发明一实施例中，根据预测结果得出至少一个增量决策树和多个模型决策树的综合性能，包括基于至少一个增量决策树和多个模型决策树的建立时间和针对增量数据的预测准确率来确定综合性能。
在本发明一实施例中，基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测，包括基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行标签预测。
在本发明一实施例中，还包括根据标签预测的结果来确定多个模型决策树和至少一个增量决策树对增量数据的预测准确率；将多个模型决策树和至少一个增量决策树的建立时间作为确定综合性能的权重，并对增量数据的预测准确率进行排序，其中建立时间长的决策树的权重小于建立时间短的决策树的权重。
在本发明一实施例中，增量决策树的数量根据模型决策树的数量确定。
在本发明一实施例中，增量决策树的数量等于模型决策树的数量的10％至30％。
在本发明一实施例中，还包括获取预定时间段内的增量数据，并基于是否存在分类模型来确定生成至少一个增量决策树的数量；其中，若存在分类模型，根据增量数据生成至少一个增量决策树。
在本发明一实施例中，还包括若不存在分类模型，根据历史数据创建包括多个模型决策树的分类模型，其中，历史数据是已分类的数据。
在本发明另一实施例中，数据处理方法包括获取预定时间段内的增量数据，并基于是否存在分类模型来确定生成决策树的数量；若存在分类模型，则根据增量数据生成增量决策树，并基于增量决策树和分类模型中的模型决策树和增量决策树来对增量数据进行标签预测，其中，增量决策树 的数量基于更新前的模型决策树的数量来确定；确定分类模型中的模型决策树和增量决策树中的各个决策树的综合性能；基于各个决策树的综合性能，从分类模型中的模型决策树和增量决策树中选取预定数量的决策树来作为更新后的分类模型中的模型决策树。
第二方面，本发明实施例还提供一种数据处理装置，该数据处理装置包括增量决策树生成模块，用于根据增量数据生成至少一个增量决策树；预测模块，用于基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测并获得预测结果；更新模块，用于根据预测结果对分类模型进行更新。
在本发明一实施例中，增量决策树生成模块包括抽样单元，用于基于增量数据有放回地抽取多个样本集；生成单元，用于基于多个样本集生成至少一个增量决策树，其中至少一个增量决策树的数量基于多个模型决策树的数量来确定。
在本发明一实施例中，更新模块包括综合性能判定单元，用于根据预测结果得出至少一个增量决策树和多个模型决策树的综合性能；更新单元，用于基于至少一个增量决策树和多个模型决策树的综合性能，从多个模型决策树和至少一个增量决策树中选取预定数量的决策树来作为更新后分类模型中的模型决策树。
在本发明一实施例中，数据处理装置包括：增量数据输入单元，被配置为获取预定时间段内的增量数据；判断单元，被配置为根据是否存在分类模型来生成表征存在分类模型的第一信号以及表征不存在分类模型的第二信号；决策树生成单元，被配置为基于响应的第一信号来根据增量数据生成增量决策树；标签预测单元，被配置为根据分类模型中的模型决策树和增量决策树来对增量数据进行标签预测；决策树选择单元，被配置为根据分类模型中的模型决策树和增量决策树中的各个决策树的综合性能来选择预定数量的决策树；模型更新单元，被配置为将经选择的预定数量的决策树作为更新后的分类模型中的模型决策树。
在本发明一实施例中，更新单元中的预定数量等于多个模型决策树的数量。
在本发明一实施例中，综合性能判定单元还用于基于至少一个增量决策树和多个模型决策树的建立时间和针对增量数据的预测准确率来确定综合性能。
在本发明一实施例中，预测模块用于基于分类模型中的多个模型决策 树和至少一个增量决策树对增量数据进行标签预测。
在本发明一实施例中，预测模块还用于根据标签预测的结果来确定多个模型决策树和至少一个增量决策树对增量数据的预测准确率；将多个模型决策树和至少一个增量决策树的建立时间作为确定综合性能的权重，并对增量数据的预测准确率进行排序，其中建立时间长的决策树的权重小于建立时间短的决策树的权重。
在本发明一实施例中，增量决策树生成模块中的至少一个增量决策树的数量根据多个模型决策树的数量确定。
在本发明一实施例中，增量决策树生成模块中的至少一个增量决策树的数量等于多个模型决策树的数量的10％至30％。
在本发明一实施例中，增量决策树生成模块还用于获取预定时间段内的增量数据，并基于是否存在分类模型来确定生成至少一个增量决策树的数量；其中，若存在分类模型，根据增量数据生成至少一个增量决策树。
在本发明一实施例中，增量决策树生成模块还用于若不存在分类模型，根据历史数据创建包括多个模型决策树的分类模型，其中，历史数据是已分类的数据。
第三方面，本发明实施例还提供一种计算机存储介质，该计算机可读存储介质上存储有数据处理程序，该数据处理程序被处理器执行时实现上述任一实施例所提及的数据处理方法的操作。
本发明实施例提供的数据处理方法通过利用增量数据进行分类模型的更新，使得分类模型能够及时地或者近似实时地根据样本数据的变化做出相应的调整，实现了分类模型与最新样本数据的同步。也就是说，本发明实施例提供的数据处理方法能够基于当前新获得的数据进行自适应更新，从而适应数据的新的趋势变化，进而保证了预测的准确性。此外，本发明实施例通过初始的操作设置，实现了在模型的业务周期内不再需要人工干预的目的，大大节约了成本，使本发明实施例提供的数据处理方法具备了智能化、高效性的特点。
附图简要说明
参考附图示出并阐明实施例。这些附图用于阐明基本原理，从而仅仅示出了对于理解基本原理必要的方面。这些附图不是按比例的。在附图中，相同的附图标记表示相似的特征。
图1所示为本发明一实施例提供的数据处理方法的流程示意图。
图2所示为本发明一实施例提供的数据处理方法的根据增量数据生成至少一个增量决策树操作的流程示意图。
图3所示为本发明一实施例提供的数据处理方法的根据预测结果对分类模型进行更新操作的流程示意图。
图4所示为本发明另一实施例提供的数据处理方法的流程示意图。
图5所示为本发明一实施例提供的数据处理装置的结构示意图。
图6所示为本发明一实施例提供的数据处理装置的增量决策树生成模块的结构示意图。
图7所示为本发明一实施例提供的数据处理装置的更新模块的结构示意图。
图8所示为本发明另一实施例提供的数据处理装置的结构示意图。
图9所示为本发明一实施例提供的数据处理装置的决策树选择单元的结构示意图。
图10所示为本发明一实施例提供的电子设备的结构示意图。
实施本发明的方式
在以下优选的实施例的具体描述中，将参考构成本发明一部分的所附的附图。所附的附图通过示例的方式示出了能够实现本发明的特定的实施例。示例的实施例并不旨在穷尽根据本发明的所有实施例。可以理解，在不偏离本发明的范围的前提下，可以利用其他实施例，也可以进行结构性或者逻辑性的修改。因此，以下的具体描述并非限制性的，且本发明的范围由所附的权利要求所限定。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论，但在适当情况下，所述技术、方法和设备应当被视为说明书的一部分。对于附图中的各单元之间的连线，仅仅是为了便于说明，其表示至少连线两端的单元是相互通信的，并非旨在限制未连线的单元之间无法通信。
发明人通过研究发现，在传统的机器学习领域，一般都是基于离线学习的机器学习模型。然而，随着数据量的增大，机器学习模型的处理能力越来越下降，特别在金融交易领域，信息瞬息万变，离线的机器学习模型会导致交易系统出现一定程度的滞后性。另外，虽然当前也存在一些基于在线学习的机器学习模型，但是复杂的结构导致工作效率较低，难以进行推广应用，尤其是难以应用在需要快速给出分析结果的金融领域。
基于上述发明构思，本发明实施例提出了基于增量数据来生成增量决 策树，然后对分类模型进行更新的技术方案。可以理解的，该增量数据可以是来自经由网络传输的金融产品信息，譬如，价格、交易金额、交易量等等。
在机器学习中，随机森林分类模型是一个包含多个决策树的分类器，并且其输出的分类结果是由单个决策树输出的分类结果的总数而定。具体来说，随机森林分类的基本思想是：从原始样本集中有放回地随机抽取N个样本集，且每个样本集的样本容量都与原始样本集一样；对N个样本集分别建立N个决策树，每个决策树都有一票选择权来选择分类结果，得到N种分类结果；根据N种分类结果对每个样本进行投票表决以决定其最终分类。随机森林生成的过程就是训练每个决策树的过程。
训练每个决策树的过程包括如下内容：(1)有放回地随机选择M个样本，用这M个样本来训练一棵决策树；(2)每个样本有多个属性，在决策树中需要分裂节点时，从这多个属性中随机选取m个属性，然后从这m个属性中采用特定的策略选择最佳属性作为当前节点的分裂属性；(3)决策树的每个节点的分裂都按照(2)进行，直到不能分裂为止。
在实际的业务应用中，获取用户行为数据后，可以先利用已在线上部署的分类模型，即由预定数量的模型决策树组成的分类模型，通过打分的方式进行类别预测，将得分最高的类别(选择该类别的决策树数量最多)作为预测类别，并基于该预测类别开展预先设定的业务应用，例如按类别进行判断价格的涨跌等。
图1所示为本发明一实施例提供的数据处理方法的流程示意图。图1的方法由服务器或计算设备执行。如图1所示，本发明实施例提供的数据处理方法包括如下内容。
11：根据增量数据生成至少一个增量决策树。
在11中，增量数据是指从数据存储设备或服务器获取的某一时间段(譬如，10分钟、1小时或1天)内的新增数据。增量决策树是一种树形结构，其中，增量决策树的每个内部节点表示一个属性测试，每个分支代表一个测试输出，每个叶节点代表一种类别。
应当理解，增量决策树中各节点所代表的属性以及类别等应根据分类模型和实际应用情况进行设定。
12：基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测，获得预测结果。
同样，模型决策树也是一种树形结构，其中，模型决策树的每个内部 节点表示一个属性测试，每个分支代表一个测试输出，每个叶节点代表一种类别。
优选地，借助标签预测方法来进行增量数据的预测操作。比如，对增量数据进行有放回抽样，抽取一定数量的样本集，然后基于抽取的样本集生成相应数量的增量决策树，最后基于生成的增量决策树对增量数据进行标签预测操作。
13：根据预测结果对分类模型进行更新。
应当理解，预测结果应当能够反映出增量决策树的综合性能，尤其是针对于增量数据的预测准确率。
在实际应用过程中，首先根据增量数据生成增量决策树，然后基于模型决策树和增量决策树对增量数据进行预测并获得预测结果，最后根据预测结果对分类模型中的模型决策树进行更新操作。
在本发明一实施例中，更新操作指的是选取综合性能较好的增量决策树来替换掉更新前的分类模型中的综合性能较差的模型决策树。
本发明实施例提供的数据处理方法通过利用增量数据生成增量决策树，然后基于分类模型中的模型决策树和增量决策树对增量数据进行预测，根据预测结果对分类模型进行更新的方式，实现了分类模型的自适应更新，并且实现了在模型的业务周期内不再需要人工干预的目的，大大节约了成本。
图2所示为本发明一实施例提供的数据处理方法的根据增量数据生成至少一个增量决策树操作的流程示意图。如图2所示，在本发明实施例提供的数据处理方法中，根据增量数据生成至少一个增量决策树(11)，包括：
21：基于增量数据有放回地抽取多个样本集。
22：基于多个样本集生成至少一个增量决策树，其中至少一个增量决策树的数量基于多个模型决策树的数量来确定。
在实际的应用过程中，首先基于增量数据有放回地抽取多个样本集，然后基于抽取的样本集生成增量决策树，其中增量决策树的数量基于模型决策树的数量来确定，然后基于分类模型中的模型决策树和增量决策树对增量数据进行预测并获得预测结果，最后根据获得的预测结果对分类模型进行更新操作。
本发明实施例提供的数据处理方法通过采用有放回地抽取多个样本集的方式来生成增量决策树，实现了使增量决策树的每个节点均选自于该样本集的特征的目的，从而为最终提高分类模型的预测精准性提供了前提条 件。
图3所示为本发明一实施例提供的数据处理方法的根据预测结果对分类模型进行更新操作的流程示意图。如图3所示，在本发明实施例提供的数据处理方法中，根据预测结果对分类模型进行更新(13)，包括：
31：根据预测结果得出至少一个增量决策树和多个模型决策树的综合性能。
应当理解，综合性能中的评判参数可根据实际情况自行设定，包括但不限于为建立时间、预测准确率等评判参数。
32：基于至少一个增量决策树和多个模型决策树的综合性能，从多个模型决策树和至少一个增量决策树中选取预定数量的决策树来作为更新后分类模型中的模型决策树。
也就是说，本发明实施例提供的数据处理方法根据各决策树的综合性能来对更新前的分类模型中的模型决策树进行选择性更新，利用综合性能较好的增量决策树来替换更新前的分类模型中的模型决策树，从而最终实现更新后的分类模型的精准预测。
图4所示为本发明另一实施例提供的数据处理方法的流程示意图。如图4所示，本发明实施例提供的数据处理方法包括如下内容。
41：获取增量数据。
在本发明一实施例中，获取增量数据指的是从金融交易服务器或特定的存储装置中获取预定时间段的增量数据。其中，该预定时间段是指位于当前时间之前的一个时间段，预定时间段的长度可以根据具体的需求进行设置，只要预定时间段内的用户行为数据已经处于可获取的状态并且已经包含了实际的类别标签信息即可。比如，预定时间段的长度预定时间段可以以天为单位、以小时为单位或以分钟为单位。
在本发明实施例中，以金融产品(比如股票)交易为例进行说明。例如，在股票交易系统中，获取距离当前时间5分钟的交易数据，数据的标签可以是上涨、下跌、平，也就是说，预定时间段为在当前时间之前的5分钟内的时间段。应当理解，在其他的实施方式中，数据的标签也可能有多种其它的形式。
42：判断是否存在线上的分类模型。
在42中，将判断是否存在能够使用的分类模型，如果存在，则执行43，否则执行49。
下面对基于分类模型是否存在不同场景分别进行阐述。
场景1：存在分类模型。
43：对增量数据进行有放回抽样，抽取k个样本集。
在43中，对获得的增量数据进行有放回抽样，生成K个训练样本集，每个样本都有类似如下所示的形式：(x 1,x 2……x n:c)，其中x i(i＝1,2……n)表示该样本的具体属性，c则表示该样本的实际类别。例如，在本实施例的一个具体例子中，在金融交易业务领域，采用分类模型对股票价格的趋势进行分类预测，每个样本的属性可以选择性地包括股票名称、价格、交易量等具体属性。
应当理解，K的具体取值可根据实际情况自行设定，以充分提高本发明实施例提供的数据处理方法的适应能力和应用广泛性，本发明实施例对此不进行统一限定。
44：基于K个样本集，创建K个增量决策树。
在44中，每个样本集生长为相应的增量决策树，即树的每个节点均是选自于该样本集的特征。
45：基于分类模型中的模型决策树和K个增量决策树对增量数据进行标签预测。
在45中，将基于分类模型中的模型决策树(假设为T个)和K个增量决策树对增量数据进行标签预测(即分类预测)，对未经分类的增量数据进行分类，如此，共有T+K个决策树对增量数据进行标签预测。由于参与预测的决策树总量的增加以及该K个增量决策树往往能代表新的趋势变化，因此利用T+K个决策树有利于提升分类模型预测的准确率。
在本发明一实施例中，为了使得新增的K个增量决策树不会损坏分类模型的准确性和适用性，设定K的取值范围为0.1T至0.3T。
注意，字母T、K仅用于表征分类模型中的模型决策树和根据增量数据生成的增量决策树的数量不同，并不旨在将T、K限定为某一具体值，例如大于或等于1的整数。
46：获得预测结果，并确定各决策树的当前准确率和建立时间。
在46中，首先将基于45中所执行的标签预测操作获得预测结果，然后将预测结果与真实的结果进行比对，从而得出各决策树的当前准确率，即针对增量数据的预测准确率。相应地，还可以获得每个决策树的建立时间，即每个决策树已经存在的时间。
47：确定各决策树的综合性能。
通过执行46，已经可以确定每个决策树的预测准确率和建立时间。在 本实施方式中，将通过该两个参数来确定每个决策树的综合性能。
在一实施方式中，综合性能＝a*建立时间+b*预测准确率，其中，a、b分别为建立时间和预测准确率的权重，a、b的取值可以根据实际情况进行调整。由此可知，决策树的生成时间对综合性能也产生影响，也就是说，最接近当前时间的决策树的权重比离当前时间相隔长的决策树的权重大。换而言之，通过对a、b值的配置，能够使得当两个决策树的预测准确率相同时，具备较短的建立时间的决策树的综合性能优于具备较长的建立时间的决策树的综合性能。
可以理解的，这里所例举出来的综合性能与建立时间、预测准确率之间的表达式仅仅旨在说明综合性能与二者相关，并非用来限定综合性能只能等于建立时间与预测准确率之和。下面结合表1对决策树综合性能的确定进行阐述。
表1 决策树综合性能
决策树ID 预测准确率 建立时间(小时) 综合性能排序
3 90％ 5 1
1 85％ 5 2
2 83％ 8 3
4 80％ 8 4
5 80％ 9 5
在本实施方式中，引入了建立时间作为影响决策树综合性能的权重。对于两个决策树的预测准确率相同的情形(比如决策树4和决策树5的预测准确率均为80％)，则进一步根据该两个决策树的建立时间来确定的该两个决策树的综合性能，即，由于决策树4的建立时间短，因此得出决策树4的综合性能优于决策树5的综合性能的结论。
48：基于决策树的综合性能，选择预定数量的决策树对分类模型进行更新。
在48中，将基于决策树的综合性能，从中选择预定数量的决策树作为更新后的分类模型的模型决策树，其中，各决策树的综合性能排序根据各决策树对增量数据的标签预测结果得出。具体而言，基于决策树的综合性能来排序，以获得表1中示出的依据综合性能排序的决策树序列，并根据 排序结果选择综合性能优秀的决策树。由前述可知，当考虑建立时间的权重时，决策树4的综合性能将优于决策树5的综合性能，因此若需要选择4个决策树丢弃1个决策树，则决策树5将被丢弃，将选择决策树1至4作为分类模型的模型决策树，更新后的分类模型将用于对后续的增量数据进行预测。
由上可知，本发明实施例提供的数据处理方法能够在保证分类模型预测准确率的前提下，实现对分类模型的更新操作。
优选地，增量决策树的数量K基于分类模型中模型决策树的数量T而确定。
在本发明一实施例中，增量决策树的数量K的范围为分类模型中的模型决策树的数量T的10％至30％。进一步地，K的具体值可以根据用户的指令或应用场景随机地在T的10％至30％之间确定，从而使得分类模型中的模型决策树的数量T也可以产生相应的变化。应当理解，本发明实施例对增量决策树的数量所进行的限定，实现了在对分类模型进行更新的情况下，不影响分类模型的稳定性的目的。
在另一种实施方式中，所选取的预定数量的决策树的数量等于分类模型中的原有的模型决策树的数量，即分类模型中的模型决策树的数量始终保持为T个，丢弃的决策树的数量等于增量决策树的数量。
为了更好地表达本发明实施例的构思，下面以T＝200，K＝40为例进行阐述。请再参考图4，在本发明实施例中，通过执行45，将使用T+K(即240)个决策树对增量数据进行标签预测，然后基于预测结果对决策树的综合性能进行排序。根据排序的结果，可以从该240个决策树中选择190、200或210个决策树来作为分类模型的模型决策树，进而完成对分类模型的更新。相应地，在下次利用该分类模型进行更新时，K可以是0.1T至0.3T中的任一数量或用户指定数量。
场景2：不存在分类模型。
继续参考图4，若42中判断为不存在可以利用的分类模型，则执行49，即基于历史数据生成模型决策树，譬如，对历史数据进行抽样，形成T个样本集，然后基于该T个样本集生成T个模型决策树。可以理解的是，历史数据是已分类的数据。
再执行410，基于49所生成的T个模型决策树构成分类模型。通过执行410，可以利用新创建的分类模型对增量数据进行标签预测(即继续执行43等后续操作)。
注意，本发明实施例没有采用基于全量数据重新构建分类模型的传统离线计算方法，而是采用增量数据进行分类模型的更新，使得分类模型能够及时地或者近似实时地根据样本数据的变化做出相应的调整，实现了分类模型与最新样本数据的同步。同时，本发明实施例通过初始的操作设置，实现了在模型的业务周期内不再需要人工干预的目的，大大节约了成本，使本发明实施例提供的数据处理方法具备了智能化、高效性的特点。
图5所示为本发明一实施例提供的数据处理装置的结构示意图。如图5所示，本发明实施例提供的数据处理装置包括：
增量决策树生成模块51，用于根据增量数据生成至少一个增量决策树。
预测模块52，用于基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测，获得预测结果。
更新模块53，用于根据预测结果对分类模型进行更新。
在本发明一实施例中，预测模块52用于基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行标签预测。
在本发明另一实施例中，预测模块52还用于根据标签预测的结果来确定多个模型决策树和至少一个增量决策树对增量数据的预测准确率；将多个模型决策树和至少一个增量决策树的建立时间作为确定综合性能的权重，并对增量数据的预测准确率进行排序，其中建立时间长的决策树的权重小于建立时间短的决策树的权重。
在本发明一实施例中，增量决策树生成模块51中的至少一个增量决策树的数量根据多个模型决策树的数量确定。
在本发明另一实施例中，增量决策树生成模块51中的至少一个增量决策树的数量等于多个模型决策树的数量的10％至30％。
在本发明另一实施例中，增量决策树生成模块51还用于获取预定时间段内的增量数据，并基于是否存在分类模型来确定生成至少一个增量决策树的数量；其中，若存在分类模型，根据增量数据生成至少一个增量决策树。
在本发明另一实施例中，增量决策树生成模块51还用于若不存在分类模型，根据历史数据创建包括多个模型决策树的分类模型，其中，历史数据是已分类的数据。
图6所示为本发明一实施例提供的数据处理装置的增量决策树生成模块的结构示意图。如图6所示，本发明实施例提供的数据处理装置的增量决策树生成模块51包括：
抽样单元61，用于基于增量数据有放回地抽取多个样本集。
生成单元62，用于基于多个样本集生成至少一个增量决策树，其中至少一个增量决策树的数量基于多个模型决策树的数量来确定。
图7所示为本发明一实施例提供的数据处理装置的更新模块的结构示意图。如图7所示，本发明实施例提供的数据处理装置的更新模块53包括：
综合性能判定单元71，用于根据预测结果得出至少一个增量决策树和多个模型决策树的综合性能。
更新单元72，用于基于至少一个增量决策树和多个模型决策树的综合性能，从多个模型决策树和至少一个增量决策树中选取预定数量的决策树来作为更新后分类模型中的模型决策树。
在本发明一实施例中，更新单元72中的预定数量等于多个模型决策树的数量。
在本发明一实施例中，综合性能判定单元71还用于基于至少一个增量决策树和多个模型决策树的建立时间和针对增量数据的预测准确率来确定综合性能。
图8所示为本发明另一实施例提供的数据处理装置的结构示意图。如图8所示，本发明实施例提供的数据处理装置包括：
增量数据输入单元81，被配置为获取预定时间段内的增量数据。
判断单元82，被配置为根据是否存在分类模型来生成表征存在分类模型的第一信号以及表征不存在分类模型的第二信号。
决策树生成单元83，被配置为基于第一信号来根据增量数据生成增量决策树。
标签预测单元84，被配置为根据分类模型中的模型决策树和增量决策树来对增量数据进行标签预测。
决策树选择单元85，被配置为根据分类模型中的模型决策树和增量决策树中的各个决策树的综合性能来选择预定数量的决策树。
模型更新单元86，被配置为将经选择的预定数量的决策树作为更新后的分类模型中的模型决策树。
由此，本发明实施例提供的数据处理装置可以在获取增量数据后，利用分类模型对该增量数据进行预测，并且还能够基于该增量数据对分类模型进行更新。也就是说，本发明实施例提供的数据处理装置实现了分类模型的自适应更新。
在一种实施方式中，决策树选择单元85所选取的预定数量的决策树的 数量等于分类模型中原有的模型决策树的数量。
在本发明一实施例中，数据处理装置还包括被配置为获取经分类的历史数据的历史数据输入单元87。具体地，当判断单元82未发现有可使用的分类模型时，决策树生成单元83基于判断单元82所生成的第二信号来根据历史数据生成模型决策树，进而生成可以使用的分类模型。
图9所示为本发明一实施例提供的数据处理装置的决策树选择单元的结构示意图。如图9所示，在本发明实施例提供的数据处理装置中，决策树选择单元85包括准确率确定单元91和决策树综合性能排序单元92，其中，准确率确定单元91被配置为根据标签预测的结果来确定各个决策树针对增量数据的预测准确率，决策树综合性能排序单元92被配置为基于各个决策树的建立时间以及对增量数据的预测准确率进行排序；其中，建立时间长的决策树的权重小于建立时间短的决策树的权重。应当理解，本发明实施例提供的数据处理装置使得分类模型能够根据数据变化的趋势来进行调整，有助于提升或保持模型的预测准确率。
应当理解，图5至图9提供的数据处理装置中的增量决策树生成模块51、预测模块52和更新模块53，以及增量决策树生成模块51中包含的抽样单元61和生成单元62，更新模块53中包含的综合性能判定单元71和更新单元72的操作和功能可以参考上述图1至图4提供的数据处理方法，为了避免重复，在此不再赘述。
图10所示为本发明一实施例提供的电子设备的结构示意图。图10提供的电子设备用于执行图1至图4的实施例中描述的数据处理方法。如图10所示，该电子设备包括处理器101、存储器102和总线103。
处理器101，用于通过总线103调用存储器102中存储的代码，以根据增量数据生成至少一个增量决策树；基于分类模型中的多个模型决策树和至少一个增量决策树对增量数据进行预测并获得预测结果；根据预测结果对分类模型进行更新。
应当理解，该电子设备包括但不限于为手机、平板电脑等电子设备。
在本发明一实施例中，还提供一种计算机存储介质，该计算机可读存储介质上存储有数据处理程序，该数据处理程序被处理器执行时实现上述任一实施例所提及的数据处理方法的操作。
应当理解，该计算机可读介质如CD-ROM、软盘、硬盘、数字通用光盘(DVD)、蓝光光盘或其它形式的存储器。替代的，图1至图4中的示例方法中的一些操作或所有操作可利用专用集成电路(ASIC)、可编程逻辑器 件(PLD)、现场可编程逻辑器件(EPLD)、离散逻辑、硬件、固件等的任意组合被实现。另外，虽然图1至图4所示的流程图描述了该数据处理方法，但可对该处理方法中的操作进行修改、删除或合并。
如上所述，可利用编码指令(如计算机可读指令)来实现图1至图4任一的示例过程，该编程指令存储于有形计算机可读介质上，如硬盘、闪存、只读存储器(ROM)、光盘(CD)、数字通用光盘(DVD)、高速缓存器、随机访问存储器(RAM)和/或任何其他存储介质，在该存储介质上信息可以存储任意时间(例如，长时间，永久地，短暂的情况，临时缓冲，和/或信息的缓存)。如在此所用的，该术语有形计算机可读介质被明确定义为包括任意类型的计算机可读存储的信号。附加地或替代地，可利用编码指令(如计算机可读指令)实现图1的示例过程，该编码指令存储于非暂时性计算机可读介质，如硬盘，闪存，只读存储器，光盘，数字通用光盘，高速缓存器，随机访问存储器和/或任何其他存储介质，在该存储介质信息可以存储任意时间(例如，长时间，永久地，短暂的情况，临时缓冲，和/或信息的缓存)。
因此，虽然参照特定的示例来描述了本发明，其中这些特定的示例仅仅旨在是示例性的，而不是对本发明进行限制，但对于本领域普通技术人员来说显而易见的是，在不脱离本发明的精神和保护范围的基础上，可以对所公开的实施例进行改变、增加或者删除。

Claims (23)

  1. 一种数据处理方法，其特征在于，包括：
    根据增量数据生成至少一个增量决策树；
    基于分类模型中的多个模型决策树和所述至少一个增量决策树对所述增量数据进行预测，获得预测结果；
    根据所述预测结果对所述分类模型进行更新。
  2. 如权利要求1所述的数据处理方法，其特征在于，所述根据增量数据生成至少一个增量决策树，包括：
    基于所述增量数据有放回地抽取多个样本集；
    基于所述多个样本集生成至少一个增量决策树，其中所述至少一个增量决策树的数量基于所述多个模型决策树的数量来确定。
  3. 如权利要求1或2所述的数据处理方法，其特征在于，所述根据所述预测结果对所述分类模型进行更新，包括：
    根据所述预测结果得出所述至少一个增量决策树和所述多个模型决策树的综合性能；
    基于所述至少一个增量决策树和所述多个模型决策树的综合性能，从所述多个模型决策树和所述至少一个增量决策树中选取预定数量的决策树来作为更新后分类模型中的模型决策树。
  4. 如权利要求3所述的数据处理方法，其特征在于，所述预定数量等于所述多个模型决策树的数量。
  5. 如权利要求3或4所述的数据处理方法，其特征在于，所述根据所述预测结果得出所述至少一个增量决策树和所述多个模型决策树的综合性能，包括：
    基于所述至少一个增量决策树和所述多个模型决策树的建立时间和针对所述增量数据的预测准确率来确定所述综合性能。
  6. 如权利要求1至5任一所述的数据处理方法，其特征在于，所述基于分类模型中的多个模型决策树和所述至少一个增量决策树对所述增量数据进行预测，包括：
    基于分类模型中的多个模型决策树和所述至少一个增量决策树对所述增量数据进行标签预测。
  7. 如权利要求6所述的数据处理方法，其特征在于，还包括：
    根据所述标签预测的结果来确定所述多个模型决策树和所述至少一个 增量决策树对所述增量数据的预测准确率；
    将所述多个模型决策树和所述至少一个增量决策树的建立时间作为确定所述综合性能的权重，并对所述增量数据的预测准确率进行排序，其中建立时间长的决策树的权重小于建立时间短的决策树的权重。
  8. 如权利要求1至7任一所述的数据处理方法，其特征在于，所述至少一个增量决策树的数量根据所述多个模型决策树的数量确定。
  9. 如权利要求8所述的数据处理方法，其特征在于，所述至少一个增量决策树的数量等于所述多个模型决策树的数量的10％至30％。
  10. 如权利要求1至9任一所述的数据处理方法，其特征在于，还包括：
    获取预定时间段内的所述增量数据，并基于是否存在所述分类模型来确定生成所述至少一个增量决策树的数量；
    其中，若存在所述分类模型，根据所述增量数据生成所述至少一个增量决策树。
  11. 如权利要求10所述的数据处理方法，其特征在于，还包括：
    若不存在所述分类模型，根据历史数据创建包括所述多个模型决策树的所述分类模型，其中，所述历史数据是已分类的数据。
  12. 一种数据处理装置，其特征在于，包括：
    增量决策树生成模块，用于根据增量数据生成至少一个增量决策树；
    预测模块，用于基于分类模型中的多个模型决策树和所述至少一个增量决策树对所述增量数据进行预测，获得预测结果；
    更新模块，用于根据所述预测结果对所述分类模型进行更新。
  13. 如权利要求12所述的数据处理装置，其特征在于，所述增量决策树生成模块包括：
    抽样单元，用于基于所述增量数据有放回地抽取多个样本集；
    生成单元，用于基于所述多个样本集生成至少一个增量决策树，其中所述至少一个增量决策树的数量基于所述多个模型决策树的数量来确定。
  14. 如权利要求12或13所述的数据处理装置，其特征在于，所述更新模块包括：
    综合性能判定单元，用于根据所述预测结果得出所述至少一个增量决策树和所述多个模型决策树的综合性能；
    更新单元，用于基于所述至少一个增量决策树和所述多个模型决策树的综合性能，从所述多个模型决策树和所述至少一个增量决策树中选取预 定数量的决策树来作为更新后分类模型中的模型决策树。
  15. 如权利要求14所述的数据处理装置，其特征在于，所述更新单元中的预定数量等于所述多个模型决策树的数量。
  16. 如权利要求14或15所述的数据处理装置，其特征在于，所述综合性能判定单元用于基于所述至少一个增量决策树和所述多个模型决策树的建立时间和针对所述增量数据的预测准确率来确定所述综合性能。
  17. 如权利要求12至16任一所述的数据处理装置，其特征在于，所述预测模块用于基于分类模型中的多个模型决策树和所述至少一个增量决策树对所述增量数据进行标签预测。
  18. 如权利要求17所述的数据处理装置，其特征在于，所述预测模块还用于根据所述标签预测的结果来确定所述多个模型决策树和所述至少一个增量决策树对所述增量数据的预测准确率；将所述多个模型决策树和所述至少一个增量决策树的建立时间作为确定所述综合性能的权重，并对所述增量数据的预测准确率进行排序，其中建立时间长的决策树的权重小于建立时间短的决策树的权重。
  19. 如权利要求12至18任一所述的数据处理装置，其特征在于，所述增量决策树生成模块中的所述至少一个增量决策树的数量根据所述多个模型决策树的数量确定。
  20. 如权利要求19所述的数据处理装置，其特征在于，所述增量决策树生成模块中的至少一个增量决策树的数量等于所述多个模型决策树的数量的10％至30％。
  21. 如权利要求12至20任一所述的数据处理装置，其特征在于，所述增量决策树生成模块还用于获取预定时间段内的所述增量数据，并基于是否存在所述分类模型来确定生成所述至少一个增量决策树的数量；其中，若存在所述分类模型，根据所述增量数据生成所述至少一个增量决策树。
  22. 如权利要求21所述的数据处理装置，其特征在于，所述增量决策树生成模块还用于若不存在所述分类模型，根据历史数据创建包括所述多个模型决策树的所述分类模型，其中，所述历史数据是已分类的数据。
  23. 一种计算机存储介质，其特征在于，所述计算机可读存储介质上存储有数据处理程序，所述数据处理程序被处理器执行时实现如权利要求1至11中任一项所述的数据处理方法的操作。
PCT/CN2018/092390 2017-06-30 2018-06-22 数据处理方法和数据处理装置 WO2019001359A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
KR1020197013526A KR20190075962A (ko) 2017-06-30 2018-06-22 데이터 처리 방법과 데이터 처리 장치
US16/362,186 US20190220710A1 (en) 2017-06-30 2019-03-22 Data processing method and data processing device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710523102.5 2017-06-30
CN201710523102.5A CN107330464A (zh) 2017-06-30 2017-06-30 数据处理方法和装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/362,186 Continuation US20190220710A1 (en) 2017-06-30 2019-03-22 Data processing method and data processing device

Publications (1)

Publication Number Publication Date
WO2019001359A1 true WO2019001359A1 (zh) 2019-01-03

Family

ID=60199340

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/092390 WO2019001359A1 (zh) 2017-06-30 2018-06-22 数据处理方法和数据处理装置

Country Status (4)

Country Link
US (1) US20190220710A1 (zh)
KR (1) KR20190075962A (zh)
CN (1) CN107330464A (zh)
WO (1) WO2019001359A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115470397A (zh) * 2021-06-10 2022-12-13 腾讯科技（深圳）有限公司 内容推荐方法、装置、计算机设备和存储介质

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330464A (zh) * 2017-06-30 2017-11-07 众安信息技术服务有限公司 数据处理方法和装置
CN108509727B (zh) * 2018-03-30 2022-04-08 深圳市智物联网络有限公司 数据建模中的模型选择处理方法及装置
CN110196792B (zh) * 2018-08-07 2022-06-14 腾讯科技（深圳）有限公司 故障预测方法、装置、计算设备及存储介质
CN110033098A (zh) * 2019-03-28 2019-07-19 阿里巴巴集团控股有限公司 在线gbdt模型学习方法及装置
CN110942338A (zh) * 2019-11-01 2020-03-31 支付宝(杭州)信息技术有限公司 一种营销赋能策略的推荐方法、装置和电子设备
CN111008119A (zh) * 2019-12-13 2020-04-14 浪潮电子信息产业股份有限公司 一种硬盘预测模型的更新方法、装置、设备及介质
CN111523908B (zh) * 2020-03-31 2023-04-07 云南省烟草质量监督检测站 一种鉴别卷烟真伪的包装机型溯源方法、装置及系统
CN112395371B (zh) * 2020-12-10 2024-05-28 深圳迅策科技有限公司 一种金融机构资产分类处理方法、装置及可读介质
CN116662815B (zh) * 2023-07-28 2023-11-10 腾讯科技（深圳）有限公司 时间预测模型的训练方法以及相关设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140172754A1 (en) * 2012-12-14 2014-06-19 International Business Machines Corporation Semi-supervised data integration model for named entity classification
CN105718490A (zh) * 2014-12-04 2016-06-29 阿里巴巴集团控股有限公司 一种用于更新分类模型的方法及装置
CN106156809A (zh) * 2015-04-24 2016-11-23 阿里巴巴集团控股有限公司 用于更新分类模型的方法及装置
US20160342755A1 (en) * 2013-06-20 2016-11-24 Microsoft Technology Licensing, Llc User Behavior Monitoring On A Computerized Device
CN106446964A (zh) * 2016-10-21 2017-02-22 河南大学 一种增量式的梯度提升决策树更新方法
CN107330464A (zh) * 2017-06-30 2017-11-07 众安信息技术服务有限公司 数据处理方法和装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140172754A1 (en) * 2012-12-14 2014-06-19 International Business Machines Corporation Semi-supervised data integration model for named entity classification
US20160342755A1 (en) * 2013-06-20 2016-11-24 Microsoft Technology Licensing, Llc User Behavior Monitoring On A Computerized Device
CN105718490A (zh) * 2014-12-04 2016-06-29 阿里巴巴集团控股有限公司 一种用于更新分类模型的方法及装置
CN106156809A (zh) * 2015-04-24 2016-11-23 阿里巴巴集团控股有限公司 用于更新分类模型的方法及装置
CN106446964A (zh) * 2016-10-21 2017-02-22 河南大学 一种增量式的梯度提升决策树更新方法
CN107330464A (zh) * 2017-06-30 2017-11-07 众安信息技术服务有限公司 数据处理方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115470397A (zh) * 2021-06-10 2022-12-13 腾讯科技（深圳）有限公司 内容推荐方法、装置、计算机设备和存储介质
CN115470397B (zh) * 2021-06-10 2024-04-05 腾讯科技（深圳）有限公司 内容推荐方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
US20190220710A1 (en) 2019-07-18
CN107330464A (zh) 2017-11-07
KR20190075962A (ko) 2019-07-01

Similar Documents

Publication Publication Date Title
WO2019001359A1 (zh) 数据处理方法和数据处理装置
US11580104B2 (en) Method, apparatus, device, and storage medium for intention recommendation
CN109816221B (zh) 项目风险决策方法、装置、计算机设备和存储介质
US9984336B2 (en) Classification rule sets creation and application to decision making
CN107871166B (zh) 针对机器学习的特征处理方法及特征处理系统
CN107609185B (zh) 用于poi的相似度计算的方法、装置、设备和计算机可读存储介质
WO2019047790A1 (zh) 生成机器学习样本的组合特征的方法及系统
CN105718490A (zh) 一种用于更新分类模型的方法及装置
TW201947510A (zh) 保險業務風險預測的處理方法、裝置及處理設備
CN107291840B (zh) 一种用户属性预测模型构建方法和装置
WO2020253038A1 (zh) 一种模型构建方法及装置
CN111582341B (zh) 用户异常操作预测方法及装置
US20190163780A1 (en) Generalized linear mixed models for improving search
CN112199374B (zh) 针对数据缺失的数据特征挖掘方法及其相关设备
CN111445139A (zh) 业务流程模拟方法及装置、存储介质、电子设备
CN108229572B (zh) 一种参数寻优方法及计算设备
CN109815475B (zh) 文本匹配方法、装置、计算设备及系统
CN110941952A (zh) 一种完善审计分析模型的方法及装置
WO2021129368A1 (zh) 一种客户类型的确定方法及装置
CN113934894A (zh) 基于指标树的数据显示方法、终端设备
CN113836244B (zh) 样本获取方法、模型训练方法、关系预测方法及装置
CN117131197B (zh) 一种招标书的需求类别处理方法、装置、设备及存储介质
US11897818B2 (en) Machine learning concrete optimization
CN111539208B (zh) 语句处理方法和装置、以及电子设备和可读存储介质
CN111221938B (zh) 一种预测结果生成方法、终端及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18824400

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 20197013526

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2019547757

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.06.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18824400

Country of ref document: EP

Kind code of ref document: A1